recommended using VBR model for CBR?, regarding http://www.hydrogenaudio.org/forums/index.php?showtopic=6536 Options

[memomai]

regarding this topic: http://www.hydrogenaudio.org/forums/index....showtopic=65367

In this article it was discussed about low bitrate CBR, that using the VBR model there are less artifacts, so qualitywise it's better ro use the VBR model in this bitrate area.

I've already asked there if it is also better to use the VBR model for 320 kbps. jmartis answered to use standard -b 320.

EDIT: is there a better CBR mode than -b XXX ?

Would it be worse to use the VBR model in CBR for bitrates like 256 kbps?

What would you use? VBR model or standard CBR model?

I'm just interested, because the development of lame is concentrated on VBR, or am I wrong?

thanks for replies.

This post has been edited by memomai: Dec 9 2008, 23:37

[halb27]

... Would it be worse to use the VBR model in CBR for bitrates like 256 kbps ... What would you use? VBR model or standard CBR model? ...

I played around with VBR -B and -b (in order to get at a higher bitrate than that of -V0) for the same reason you consider it interesting.
I found however that the higher nominal bitrate consists more or less of plain air. After applying mp3packer I got at the same average bitrate as when using -V0. So you can use -V0 right away.

I played with very high bitrate ABR like ABR 280 or 290. ABR behavior here is a bit strange too. The average nominal bitrate gets way beyond 280 resp. 290 kbps, and it's an air stream too. After applying mp3packer I got at ~270 kbps, and I was very content with the quality. When out for best quality using Lame I'd go with this. Quality using -V0 or -V0.5 however is great too.

This post has been edited by halb27: Dec 8 2008, 13:29

[greynol]

Please, let's do without the subjective nonsense.

I can tell you with 100% certainty that -V0 -b320 gives a different playback stream from -V0. I'm not about to comment on the differences in quality, and neither should halb27. "Best quality" is in the ear of the beholder and neither of us have any business telling people otherwise.

This post has been edited by greynol: Dec 8 2008, 19:54

[[JAZ]]

Please, let's do without the subjective nonsense.

Haven't you jumped the gun a bit too fast? I don't see halb27 as an erroneous reply, neither I see an answer in yours.

It is true that the bitstream will be different, just for the sake that one will be variable in bitrate and the other will not.

What halb27 has pointed out is that the encoder is asked to produce the same quality ( -V 0 ) so it will reach that quality with or without the -b restriction. As per halb27 comment, what the encoder is putting there (to "fill the extra space") is mostly padding.

halb27 answer would imply that using the VBR model for CBR would produce not much difference than using the -V 0 setting alone. As thus, implying that if plain CBR 320 is better than -V 0, then using the VBR model for creating a constant bitrate 320kbps file would be worse.

Where is your conclusion?

[greynol]

Haven't you jumped the gun a bit too fast?

No, not in the least.

It is true that the bitstream will be different, just for the sake that one will be variable in bitrate and the other will not.

I'm talking about the decoded output.

As per halb27 comment, what the encoder is putting there (to "fill the extra space") is mostly padding.

Your proof?

halb27 answer would imply that using the VBR model for CBR would produce not much difference than using the -V 0 setting alone.

This implication is erroneous.

Where is your conclusion?

I'm not interested in making subjective conclusions. I guess you missed my point.

This post has been edited by greynol: Dec 8 2008, 21:10

[[JAZ]]

I'm talking about the decoded output.

Ok, verified this personally. The decoded output is not bit identical. The difference is also bigger than one bit.

Your proof?

-V 0:
kbps LR MS % long switch short %
233.9 88.3 11.7 67.9 13.6 18.5
-V 0 -b 320:
kbps LR MS % long switch short %
319.5 88.3 11.7 67.9 13.6 18.5

-V 0.5:
kbps LR MS % long switch short %
223.7 87.1 12.9 67.9 13.6 18.5

-b 320 (CBR):
320.0 88.8 11.2 78.5 10.6 10.9

The encoder is applying the same parameters. It is just forced to use more bitrate. The only theorization/subjective coclusion that we reached was about what the bitrate is used for.
This is the result of the concrete encoding I've done:

-V0 : 234kbps
-V0 -b 320: 319kbps
-b 320: 320kbps

-V0 repacked: 217kbps
-V0 -b 320 repk: 227kbps
-b 320 repakd: 305kbps

In other words. Most of the -V0 -b 320 setting became padding. Not all (I never said all).
And as you see, it is much more different to -b 320 than to -V 0. (I guess it would be an expected result, but worth noting it).

halb27 answer would imply that using the VBR model for CBR would produce not much difference than using the -V 0 setting alone.

This implication is erroneous.

The implication is correct. Don't miss the "not much".

Where is your conclusion?

I'm not interested in making subjective conclusions. I guess you missed my point.

You were presented a test, albeit incomplete. You refused to talk about it, because it was not a solid fact.

This post has been edited by [JAZ]: Dec 8 2008, 21:50

[greynol]

How do you know how much is "not much"? Are you comfortable enough to speak for all listeners and all samples?

[[JAZ]]

Just like an ABC-HR test for a single user and sample proves nothing, I don't imply that what I've said can be written in stone.
What I have said is that it is legitimate to talk about this.

I am confident that for several MP3 created with -V 0 and -V 0 -b 320, after mp3packing them, the difference in size won't be big.
Since the encoder is the same, and the internal settings are almost the same, I am confident that the result of the decoded audio won't be much different.

You ask "how much is not much?" Well, not much is less than the difference to CBR 320. Can I quantify it? No I cannot.



Edit: Just to conclude my implications:

* -V 0 -b 320 kbps is of higher quality than -V 0
* If using -V 0 -b 320, it is highly recommended to use mp3packer afterwards.
* If, for a given sample, CBR 320 is of higher quality than -V0, chances are high that CBR 320 will still be of higher quality than -V 0 -b 320.

This post has been edited by [JAZ]: Dec 8 2008, 22:29

[greynol]

What I have said is that it is legitimate to talk about this.

...and I have never said otherwise. All I ask is that the subjectivity be held to a minimum. Likening the difference to "air" is BS and I'd be remiss not to call it out as such.

The question at hand is whether -V0 -b XXX -B XXX produces better quality than -b XXX or any other method of creating CBR files with Lame. What we need are samples, not useless talk about "air" and OT soapbox preaching about ABR 280. Yes, I know this is extremely difficult with bitrates that typically give transparent results.

* -V 0 -b 320 kbps is of higher quality than -V 0

Higher bitrate does not imply higher quality.

* If using -V 0 -b 320, it is highly recommended to use mp3packer afterwards.

This is irrelevant to the topic.

* If, for a given sample, CBR 320 is of higher quality than -V0, chances are high that CBR 320 will still be of higher quality than -V 0 -b 320.

The key word here is "if" and it is at the very crux of this discussion.

http://www.hydrogenaudio.org/forums/index....st&p=573173

This post has been edited by greynol: Dec 9 2008, 00:25

[halb27]

Everybody can do as he pleases.

As for me: if a certain setting of certain encoder yields a bitrate which is not used for audio content to a large extent to me there must be a very good reason to use this setting. Beyond me I'd call it common sense.

As far as quality is concerned I did listening tests (sure restricted), and based on these to me there is no reason to use something like -V0 -b xxx. From my listening tests I can suggest to use ABR 280 or similar, and to use mp3packer afterwards in order to save some file size.

Sure I'm subjective as I care about those samples which are interesting to me.
I can't see what's wrong calling a bit stream having air in it when a significant amount of data doesn't carry audio information.

After all the OP wanted an advice on -V0 -b xxx -B yyy, and I gave it. I didn't say this is the entire truth, and everybody is welcome to add experience. But this must be done. Just calling real experience subjective isn't helpful. So far the things I mentioned and for which [JAZ] gave details are the most significant contributions towards the OP's question.

This post has been edited by halb27: Dec 8 2008, 23:33

[Canar]

What would you use? VBR model or standard CBR model?

Standard CBR model. LAME development is focused on creating an exceptional product. To trivialize their efforts and make claims like LAME focuses on VBR is to completely disregard the effort put into other areas. Yes, there might be effects to using weird switches with CBR encoding. And so, unless you're a developer, you probably should not use any more switches than you need.

On the other hand, if you really want to know which is better, run some blind tests. Your subjective experiences will be welcomed when backed by blind test results.

[greynol]

@halb27:
If I said the difference between -V4 and -V5 was only air because I could not distinguish a difference, you might feel the same way. Your suggestion was no different.

So far the things I mentioned and for which [JAZ] gave details are the most significant contributions towards the OP's question.

As it seems that you don't really know why this thread got started I would suggest you have little basis to decide what is a significant contribution to the discussion.

@Canar:
Please check the links that were given. In the case of CBR 96, you'll see that it was indeed a developer who suggested trying the VBR model. You'll also see that using the VBR model produced a superior result.

I'd love to see a constructive conversation on the topic rather than complete avoidance by way of offering off-topic alternatives and subjective hand-waving regarding differences in bitrate.

This post has been edited by greynol: Dec 9 2008, 00:27

[halb27]

a) the OP wanted to take the idea of the thread he mentioned (CBR 128 supposed to be inadequately inferior to a corresponding VBR setting) over to the very high bitrate range of 256 kbps or more and asked for advice.

b) [JAZ] and I showed that when doing so most of the extra data amount compared to the data amount of -V0 doesn't contain audio content. This large amount of non-audio content is wasted in a very objective sense (something I called plain air). And according to my own tests with numerous tracks nearly everything of the extra amount of data is wasted.
This doesn't tell about audio quality though.

c) Everybody can decide on his own but I can't imagine many people find it reasonable to objectively waste most of the extra file size/bitrate compared to that of -V0 for the mere hope that quality may increase due to the hardly increased amount of audio data. And my own (restricted) listening tests have shown me that I get what I can expect for: nothing. Different experience welcome.

d) As an alternative to the non-attractive VBR -b xxx setting in the very high bitrate range I suggested to use very high bitrate ABR (which btw is a variable bitrate method too), followed by mp3packer in order to squeeze out the air which very high bitrate ABR has as well. I did listening tests using Lame 3.98.2 --abr 285 --lowpass 16.5 and was very happy with the result. If out for very high bitrate beyond that of -V0, but below that of -b 320, this or a similar setting may be worth giving it a try. BTW Lame 3.98.2 --abr 285 --lowpass 16.5 yields an average bitrate of 268 kbps after mp3packing with my (for me) representative test set of pop music of various kind. It was my favorite setting before I decided after the last mp3 listening test that I can be content with a lower bitrate.

e) What is not so clear is whether the OP looks for using extremely high bitrate higher than that of -V0 for the sake of achieving a hopefully better quality than -V0 gives. This is what I have ín mind with my answers. Another possibility is that he is obliged to use a constant high bitrate like CBR 256 or CBR 320 and looks for the best way of doing it. My practical answers do not apply to this, but they give some facts about the large amount of wasted non-audio data when using -V0 -b xxx -B xxx, with xxx=256 or 320.

Was there any other answer to the OP's question in the thread? I don't think so. So these things are not only significant contribution, they are - apart from Canar's warning against special settings - the only contribution so far. More contribution welcome, but please no laid-back non-content talk even though by a super moderator.

This post has been edited by halb27: Dec 9 2008, 13:48

[greynol]

Another possibility is that he is obliged to use a constant high bitrate like CBR 256 or CBR 320 and looks for the best way of doing it. My practical answers do not apply to this, but they give some facts about the large amount of wasted non-audio data when using -V0 -b xxx -B xxx, with xxx=256 or 320.

This is precisely what the discussion is about and as you admitted yourself, these "facts" don't "tell about audio quality". Considering this, most of what you've said is not on-topic, nor particularly useful in answering the question.

Canar's warning against special settings - the only contribution so far.

Canar was not familiar with the history at the time he made that comment.

Further talk about ABR and subjective evaluation about sound quality not supported by positive blind test results will be removed from this discussion.

[halb27]

What is not so clear is whether the OP looks for using extremely high bitrate higher than that of -V0 for the sake of achieving a hopefully better quality than -V0 gives. ...

Another possibility is that he is obliged to use a constant high bitrate like CBR 256 or CBR 320 and looks for the best way of doing it. .....

This is precisely what the discussion is about ...

@memomai:
Would you please be so kind to make things clear in this respect?

This post has been edited by halb27: Dec 9 2008, 16:51

[Canar]

Read the previous thread, halb27. It becomes pretty clear then. greynol was exactly right, I was missing that context.

With some testing by someone with good ears, using VBR modes for CBR could give a significant quality increase with minimal effort. That's what the discussion is about. Unfortunately, that possibility is on hold while we wait for someone with good ears to try do some testing. Perhaps if I can find a bit of time I'll see what I can do... there aren't many samples that make 320kbps easy for me to ABX though.

This post has been edited by Canar: Dec 9 2008, 18:37

[halb27]

Read the previous thread, halb27. It becomes pretty clear then. ...

I read it and re-read it right now. The thread itself isn't really clear about it, in part it's a thread like '3.98 CBR could do better', but it has other aspects as well. Clear is only memomai's question there about - b 320 vs. -V0 -b 320. But in this thread here he put the question a bit wider.
When it's about CBR 320 it's usually about using best expected quality, not about being obliged to use constant bitrate. This needn't be true for memomai, but only memomai can tell. Please do, memomai.

I enjoy the great progress VBR has made with 3.98. But I can't understand VBR lovers over-generalizing the advantages of VBR. In my understanding everybody who sees the extreme amount of garbage data when using -V0 -b 320 (and it takes only to use the lossless mp3packer and look at the average bitrate afterwards) can't really consider to use this setting as long as there is not the least evidence for this to be useful.
This is not a general statement on -Vn -b xxx vs. -b xxx. Things are different at low/moderate bitrate. But at very high bitrate considered here it's like that.

This post has been edited by halb27: Dec 9 2008, 18:58

[halb27]

Perhaps if I can find a bit of time I'll see what I can do... there aren't many samples that make 320kbps easy for me to ABX though.

Oh, apart from typical pre-echo samples (for which I wouldn't expect a real improvement over -V0 as -V0 is great here within the mp3 restrictions) I can suggest to use harp40_1 for a hard harpsichord problem, and herding_calls as well as trumpet for tonal problems. -V0 isn't totally fine with these samples (though quite alright), so there's room for sporting for better quality. I'd really like to know for these samples whether you can outperform the quality of --abr 285 --lowpass 16.5 (yielding 268 kbps on average for pop music after mp3repacking) by a VBR setting. Maybe -b 320 does, but at a significantly higher bitrate.

This post has been edited by halb27: Dec 9 2008, 19:11

[greynol]

No one is over-generalizing the advantages of VBR in this thread, nor what has led up to it. What we have seen are signs of regression of CBR since at least as far back as 3.96 and are interested in knowing whether the new VBR model can produce better results.

In my understanding everybody who sees the extreme amount of garbage data when using -V0 -b 320

Let me suggest that you keep your "understanding" to matters that consider yourself.

and it takes only to use the lossless mp3packer and look at the average bitrate afterwards

...which (again!) says nothing about the quality.

can't really consider to use this setting as long as there is not the least evidence for this to be useful.

Why not let it play out instead of pretending like there is no possibility that there may be evidence?

Speaking only for myself, I don't trust your ears nor your understanding of the subject, halb27.

This post has been edited by greynol: Dec 9 2008, 20:00

[Slipstreem]

I may be wrong, but memomai's main point of interest seems to be...

Would it be worse to use the VBR model in CBR for bitrates like 256 kbps?

...to which I would answer, why use 256Kbps CBR in the first place unless you're forced to by hardware incompatibility with full-blown VBR?

Would it not be worse than just using, say, VBR at -V2 because of the bitrate being capped at 256Kbps instead of being free to hit 320Kbps when the encoder sees fit, or is that just too simplistic a viewpoint?

Looking at it from my point of view, there seem to be two ways of looking at this specific question. One is to see it as a possible improvement over conventional 256Kbps CBR, and the other is to see it as a deliberate strangulation of VBR. Any thoughts?

Cheers, Slipstreem.

[greynol]

Maybe -b 320 does, but at a significantly higher bitrate.

...yet both you and [JAZ] made it perfectly clear that the bitrate can be reduced through repacking as well.

Looking at it from my point of view, there seem to be two ways of looking at this specific question. One is to see it as a possible improvement over conventional 256Kbps CBR, and the other is to see it as a deliberate strangulation of VBR. Any thoughts?

Although I am certain that the OP is interested in the former, only he can tell us.

[halb27]

... Speaking only for myself, I don't trust your ears or your understanding of the subject, halb27.

Of course you needn't, but so far it's the only contribution to the subject based on experience, no matter how questionable to you personally.

... instead of pretending like there is no possibility that there may be evidence ...

Oh no, take my posts as a challenge for -V0 -b xxx lovers to come up with an evidence that -V0 -b xxx, xxx=256 or 320, is useful.

... ...yet both you and [JAZ] made it perfectly clear that the bitrate (of -b 320) can be reduced through repacking as well.

Not by much, but -b 320 may be a good solution. If it were only about -b 320 vs. -V0 -b 320 I'd pick -b 320 like jmartis suggested.


I won't argue with you any further, greynol. Everything that can be said right now was said.

This post has been edited by halb27: Dec 9 2008, 20:00

[greynol]

Scratch my comment about repacking -b 320. It was made in error.

[[JAZ]]

Nice to see the conversation went forward.

I had read that thread about CBR doing a bit worse on the lower end of the bitrate range. That thread could be considered as a regression with one setting and version. As such, not extrapolable to general recommendations.

I wanted to keep this thread (and my replies) not restricted to a concrete version, since that would do worse that otherwise.


What we had here was the possibility that the CBR model would do worse (even though needing more bitrate) than the VBR model. The general rule has always been that -b 320 (CBR) is the best that MP3 can do. It always has been suggested it to be as much good or better than -V 0. (more the latter than the former).
This, for a general rule, is a good thing to say.



But let's go back to LAME 3.98.x VBR and CBR models.

To give an oppinion on how would these findings on lower bitrates translate to higher bitrates, first we have to understand what has happened.

The findings on that thread suggested that the CBR model was trying to keep too much quality on something it couldn't, and the lack of bitrate to keep that information caused worse artifacts than those that existed before.

Is this gradually changing when going up, and as such, pretending to keep way too much quality at 320kbps? I don't know. I don't have material to test it, but I would expect to be doing better here.

About the VBR model, keeping the bitrate up doesn't do miracles. As an example:

-V 9
-V 9 -b 320 (Actually, the limit is 160kbps, since V9 is Mpeg II)

These two, encoded, produce the same audible quality.

And also, once processed with mp3packer, they even have the same bitrate on the samples i've tested.

[memomai]

QUOTE(halb27 @ Dec 9 2008, 12:24)
What is not so clear is whether the OP looks for using extremely high bitrate higher than that of -V0 for the sake of achieving a hopefully better quality than -V0 gives. ...
QUOTE(greynol @ Dec 9 2008, 17:14)

Also an interesting point, but this wasn't my question.

QUOTE(halb27 @ Dec 9 2008, 02:24)
Another possibility is that he is obliged to use a constant high bitrate like CBR 256 or CBR 320 and looks for the best way of doing it. .....
This is precisely what the discussion is about ...

Exactly what I'm looking for: is there a better CBR mode than -b XXX (I'll edit this to the start)?

=> Do I achieve a quality gain in using the VBR model for high bitrate CBR (it's just the question if there is a gain, no matter how big it is), when the VBR model brings quality gain for low bitrate (shown in prevoius topic, see link in topic description)?

@memomai:
Would you please be so kind to make things clear in this respect?

Example: has "-V0 -b 256 -B 256 (and following paramters for forcing CBR header and so on)" got better quality than -b 256?

where 256 is just an example for the high bitrate range.

I'd like to leave out the question of compatibility first (= does "-V0 -b256 -B256" give me the same overall compatibility, e.g. streaming, like -b 256?)

These two, encoded, produce the same audible quality.

@[JAZ]: So is the conclusion right: if -V 0 brings worse quality than -b 256, then -V0 -b256 -B256 brings worse quality than -b 256?